In Which Kana is Delighted by a New Toy

I’ve been playing.

It was another blogger, The AI Grandad, whose post about “falling down the AI rabbit hole” led me to exploring MidJourney, an AI program that produces images from the written prompts you give it. This rabbit hole runs deep!

For a start, I managed to create a pretty good approximation of Yoda-cat’s “bonkers face,” which (as I’d just written) continues to defy photography.

I played around with the idea of a vintage travel poster showing someone stepping out of a book and onto a seashore.

I played around with some self-portraiture…

…and a portrait of a marriage (you may be sensing a thematic pattern here)…

…and portraits of our sailboat.

I played with drawings of characters in my novel.

I created an image of creating my novel.

And then I got interested in people’s process with this tool.

As I scrolled through the gallery of images and video clips that other people had created with MidJourney, I discovered that I could see what they had written to arrive at that image. (To arrive at many images, in truth. The program creates four images at a time, from which you can select one and request subtle changes or drastic ones, also four at a time. You can adjust what you’ve written, or edit specific details, or just keep requesting changes and see what grows out of that start. But if you enter the exact same text a second time, you’ll come up with entirely different images.)

What intrigued me was the depth of detail and the creativity of the writing in many of these prompts. I had only written the physical/visual description of what I wanted, although I got more detailed as I went along and got into the process. But what stood out, in other people’s prompts, was how they evoked not only a physical description, but a mood.

That struck me as mildly ironic, to be setting a “mood” for an entity that can’t feel. But essentially, people were writing as if it were a page of a novel, and getting the image from the mood they had set, with the details they specified. Sometimes the writing was good—as in, these were novels I would read.

The translucent rabbit sprints through an infinite mirrored garden made of pills and sound waves. Each step fractures reality glass grass bends, sky folds in half, and new rabbits burst from the reflections like echoes of thought. Thousands multiply until the whole landscape becomes a ribbon of translucent motion. Pill-trees twist and coil like serpents of light, their petals opening into eyes, watching the scene unfold. The clouds melt into glitching static and leak down like liquid code. Colors spill over: orange bleeds into violet, violet into teal, teal into black. The ground becomes water, then vapor, then a pulsating grid of neurons. The rabbits begin to run upward, defying gravity, looping through floating orbs that hum faint melodies. The horizon collapses and rebuilds itself a kaleidoscope of pill shapes, breathing textures, and vapor trails. The air vibrates with faint whispers and strobing reflections.

Frankly, some of these put my bare physical descriptions to shame.

You can also animate your creations, and some users wrote as if they were film directors—setting up the shot and the angles, specifying the lighting, painting the scenery. (One even wrote of “casting” the people in the shot.)

Ultra-wide forced-perspective cel animation. Front POV inches from the motorcycle front tire, lens ~14-16mm look. Road stripes and speed lines stretch past camera, lamp posts and cliffs warp outward, coins/obstacles fly by huge in foreground, intense parallax and squash-and-stretch on suspension. Thick clean outlines, flat fills, minimal shading, crisp vector look, stepped 12 fps animation. Rider stylized, no text. colors of desert dusk: sand, terracotta, sage, dusk purple, deep aubergine

I was entertained by some users who addressed the AI as if there were actual people behind the process. “Can you make this…?” “Let’s remember we need …” “Imagine an …” “Are you able to …?” And, touchingly, someone addressed the AI with “please.” (Actually I applaud that. I’ve wondered if there’s a generation growing up with the habit of commands and demands aimed at Alexa & Siri, who will forget their “please” and “thank you” in the world of real people…)

Then there was the person who framed his instructions as a mission statement, assigning the AI generator the role of… AI generator.

# You are an AI 3D icon generator. ## Purpose Generate 3D icons in the exact style shown in the attached showcase image … # Current Task Generate an icon that would symbolize an opportunity for a bank loan product that is handled and delivered by an AI agent. So user will discuss his loan options and preferences with AI agent in this opportunity.

Some people were playing with abstract concepts to see what images would come out of them. I saw prompts for “premonition,” “reincarnation,” “angel numbers,” and (my favorite of this type) “Her hair is the multiverse.” One person wrote a Diwali prayer. No overt instruction, just the prayer.

What all of my snooping brings to the forefront is that the AI generator is a TOOL, being used by creative HUMANS. It’s simply another outlet for human creativity. I emphasize this point in particular because of a discussion that landed on me yesterday evening, with a friend who—how shall I put this?—a friend who believes the content of different news outlets than I do. (That statement may be the closest I ever come to discussing politics here.)

I hadn’t mentioned what I’d been wrapped up in all afternoon, but she brought up AI, in the context of an imminent existential threat to humankind. You know, the “Terminator” scenario where machines have taken over the world and decide to obliterate humankind.

Suffice it to say I’m not losing sleep on this. Alarmists ascribe motivations to machines, as if machines had the same survival instinct that’s hard-wired into living organisms. As for me, I’m perfectly happy to let ChatGPT comprehensively answer the questions I pose to Google, and to let Alexa add to my grocery list, and to let MidJourney restyle my writing into a visual. (Heck, for that matter I’m happy to let Google do Google for me—it’s so familiar to us, at this point, that my friend probably doesn’t even consider it a form of AI.)

In any case, what AI can’t do is BE ME. It can’t think of the various things in my day that connect to other things that surface in a blog post. I wouldn’t want to ask AI to “guest post” for me (unless it were in the nature of an experimental venture, for fun), because I want to write! I get to write. And today, I’m enjoying what my writing can do in conjunction with a new toy tool. I’m playing.